What Do Named AI Agents Mean for Small B2B Teams?

What Do Named AI Agents Mean for Small B2B Teams?

As of September 2026, the big platforms have started shipping AI agents with human first names and job titles. For a small B2B team, the naming is mostly marketing. What matters underneath is narrower: a set of tasks a vendor is now confident enough to sell as a role rather than a feature.

We build automations for teams of ten to two hundred people, not for enterprises with a platform team. So when a vendor announces a workforce of named agents, the useful question is not whether it is impressive. It is which part of it a small team can copy without buying the platform.

Here is what was actually announced, what the published numbers do and do not support, and what we would build instead.

What Did Salesforce Actually Announce?

On 11 September 2026, Salesforce introduced a portfolio of what it calls job-ready agents, built to take on high-value work across sales, service, commerce, employee experience and the back office, connected to Customer 360. The agents have names and defined jobs rather than being configured from a blank canvas.

The named lineup includes Casey for customer service help, Paige for IT and HR service, Carter as a shopper agent, Hunter for outbound sales, Marshall for supply chain, and Piper for inbound pipeline generation. The framing is deliberate: these are described as roles a company hires rather than tools a company configures.

This sits on top of the wider Agentforce 360 release, which Salesforce describes as introducing a new Agentforce Builder, an Agent Script language for controlling agent behaviour, Agentforce Voice, and Intelligent Context for grounding agents in unstructured data.

Why Are Vendors Giving Agents Human Names?

Because a name makes the buying decision easier. "Should we hire an outbound sales agent called Hunter" is a question a sales leader can answer. "Should we configure an LLM workflow with tool access to our CRM" is a question that goes to a committee and dies there.

That is not a criticism. Naming a capability is how software has always been sold, and a clear job description forces the vendor to be specific about scope, which is genuinely useful. An agent with a name has an implied set of things it will not do.

The risk is that the name does the thinking for you. A person with a job title comes with judgment, accountability and the ability to notice that the task is wrong. An agent with a job title comes with a scope document. Those are not the same purchase, and the naming quietly encourages you to treat them as though they are.

What Do the Published Numbers Really Say?

They are customer-reported results for specific deployments, and they are worth reading precisely. Salesforce's announcement says 50 percent of Engine's chat inquiries are fully resolved by its help agent, Eva, and that 60 percent of Perk's sales pipeline is built by its outbound sales agent, Hunter.

It also cites 70 percent of Autism Queensland's administrative requests resolved by its employee service agent, 90 percent of core shopper journeys handled by Hibbett AI, and four times the conversation volume driven by Asana's website agent, Piper.

Every one of those is a single company's number, published by the vendor, with no baseline and no measurement window stated. They tell you the ceiling is real for someone. They do not tell you what a company with your data quality and your volume should expect, and treating them as a forecast is how automation budgets get set wrong.

What Is an Agentic Work Unit and Should You Care?

It is Salesforce's own unit for counting agent activity, and the company says it delivered 7 billion Agentic Work Units across Agentforce and Slack, including 3.2 billion in the second quarter alone. That is a usage metric, not an outcome metric.

Vendor-defined units are worth watching for a specific reason: they usually become the billing unit. When a platform starts counting work in a unit it invented, the shape of your future invoice is being defined in public.

For a small team the practical advice is simple. Before adopting anything priced this way, work out what one useful outcome costs in the vendor's units, not what a thousand units cost. We went through that kind of cost modelling in our piece on controlling AI costs on web teams.

Does Any of This Transfer to a Twenty Person Company?

The task definitions transfer. The architecture mostly does not. The genuinely useful thing in a named agent lineup is that someone has done the work of deciding which jobs are narrow enough for an agent to do reliably, and published the list.

Inbound pipeline qualification, first-line help, and routing internal requests are on that list because they share a shape: high volume, low individual stakes, clear success criteria, and a human available when it goes wrong. That shape is the actual lesson, and you can apply it with tools you already pay for.

What does not transfer is the assumption of a platform underneath. Those deployments sit on years of structured CRM data. If your customer data lives in three spreadsheets and an inbox, the agent is not your first problem.

Where Do Named Agents Genuinely Help?

Where a person is currently doing a repetitive job badly because they are bored. First-line support triage, chasing missing form fields, and qualifying inbound leads all fit, and a website agent handling inbound questions is now a well-trodden path rather than an experiment.

They also help where response time matters more than polish. An answer in ten seconds that is 80 percent right often beats a perfect answer tomorrow, particularly for inbound enquiries where the buyer is comparing three vendors in one sitting.

We wrote about the qualification side of this in our piece on AI lead qualification in the CRM, and the rules there hold whether the agent has a name or not.

What Breaks When You Personify an Automation?

Accountability gets fuzzy. When a workflow fails, someone owns the workflow. When Hunter fails, the conversation drifts into whether Hunter is underperforming, which is a category error that wastes a month.

The second failure is scope creep by analogy. A named agent invites people to ask it things a person with that job would handle, which is a much larger set than the tasks it was built for. The requests arrive politely and the system quietly starts operating outside what it was tested on.

The third is trust transfer. Customers who believe they are talking to a colleague extend a colleague's benefit of the doubt. That is fine while it works and expensive when it does not, which is an argument for being clear about what someone is talking to.

What Would We Build Instead?

One narrow automation, named after its job rather than after a person, with a logged handoff to a human and an owner on the team. Start with the highest volume, lowest stakes task you can find, and measure the handoff rate rather than the resolution rate.

The handoff rate is the honest number. Resolution rate flatters a system that quietly gives unhelpful answers people give up on. Handoff rate plus what happened after the handoff tells you whether the automation is saving time or moving it.

If you are weighing one agent against several, we covered the trade-offs in one agent versus many agents. For most small teams the answer stays one, for longer than the market implies.

Where Does This Land in a Year?

Our guess is that the names fade and the task catalogues stay. Vendors are currently competing on how human their agents sound, which is a phase every interface goes through, and it usually ends when buyers start comparing outcomes instead of demos.

What will not fade is the underlying shift: routine inbound work is being absorbed by software faster than most teams have planned for, and the companies that benefit are the ones whose data was already in decent shape. That is an unglamorous advantage and it is the one worth building this year.

If you want help working out which of your inbound tasks are actually automatable, and which are just annoying, we are happy to walk through it. You can reach our team at phoenix.studio.